Realism

Judges were able to tell real and generated sound effects appart 70% of the time according to our results (Fig. \ref{850789}). However, individually three out of the five judges were only able to achieve 60% accuracy. This indicates that while sometimes our model produces sound effects that are obviously generated, the majority of the time it generates sound effects that are very similar to those in the original dataset. Again we note that generated fire sound effects are usually indistinguishable from the real fire sound effects.